Federal lawmakers plan to introduce a bill Thursday that would ban the use of DeepSeek from all U.S. government devices.
The “No DeepSeek on Government Devices Act” would require the Office of Management and Budget to develop guidelines within 60 days for removal of DeepSeek’s artificial intelligence (AI) app from federal technologies, with exceptions for law enforcement and national security-related activity.
The bipartisan legislation, from Reps. Josh Gottheimer, D-N.J., and Darin LaHood, R-Ill., addresses growing security concerns that the app could be relaying personal information to the Chinese government.
The proposed ban, first reported by the Wall Street Journal, comes after an ABC News dispatch this week that revealed DeepSeek’s app may contain intentionally hidden code capable of transferring user data to an online registry for China Mobile, a telecommunications company owned and operated by the Chinese government.
“The technology race with the Chinese Communist Party (CCP) is not one the United States can afford to lose,” LaHood said in a statement. “The national security threat that DeepSeek — a CCP-affiliated company — poses to the United States is alarming.”
The contours of the bill are similar to a federal law that prohibited the use of Chinese-controlled TikTok devices from government devices which ultimately led to a national ban of the popular short-video app for 24 hours. Gottheimer is one of the lawmakers behind the TikTok bill.
Sean Tufts, managing partner for critical infrastructure and operational technology at Optiv, said the idea of DeepSeek developing comprehensive profiles of U.S. citizens, especially government officials, is “deeply concerning.”
“We simply cannot risk funneling sensitive information and government secrets directly to the government of China,” Tufts said.
“Banning DeepSeek is a good first step, but if we are to be serious about cybersecurity, we need to start the lengthy process of validating the legitimacy of all software and hardware used in the federal government and critical infrastructure space,” Puglia said.
Security concerns have dogged DeepSeek since the Chinese AI startup jolted the market with its R1 reasoning model in late January. While gloating about its benchmark tests vs. rivals OpenAI, Anthropic, Meta Platforms Inc., and others – all at a fraction of cost of their training costs – DeepSeek has proved to be problematic for security, regulatory compliance, and the broader geopolitical context associated with AI models linked to China.
“Imagine someone uploading a file to [DeepSeek] and asking what is in the file — if it contains sensitive information,” Chris Mattmann, chief data and AI officer at UCLA, said in an email.
“It was reported over the last few weeks that government employees, including in the Department of Defense, were using the DeepSeek AI service to answer questions. A query like, ‘I’m flying to Estonia, what do I need for travel?’ could tip off the Chinese to U.S. Department of Defense activities,” Daron Hartvigsen, managing director at StoneTurn and a former Special Agent and Counterintelligence Investigator with the U.S. Air Force Office of Special Investigations, said in an email.
“The current and prior administration have made it clear that the U.S. is aiming to stay one step ahead of China with respect to innovation and maturation of these technologies, so it is a logical step to thwart its evolution at the expense of U.S. intellectual property theft,” Hartvigsen said.
Still, DeepSeek is already prohibited in Italy, Ireland, Australia, and by the U.S. Navy and Congressional office workers.
Organizations that haven’t blocked DeepSeek are applying strict internal policies, third-party risk assessments, or controlled environments to balance innovation with security considerations, according to experts.
“This reflects a broader trend of enterprises applying increased scrutiny to all AI models, regardless of origin, to ensure alignment with their own internal cybersecurity and compliance frameworks,” Kelcey Morgan, senior manager of orchestration, intelligence and AI at Rapid7, said in an email.
